A machine learning model for estimating daily maximum 8-hour average ozone concentrations using OMI and MODIS products.
In: Atmospheric Environment, Jg. 331 (2024-08-15), S. N.PAG
academicJournal
Zugriff:
Tropospheric ozone (O 3) is a criteria air pollutants posing risks to organisms, and is expected to enhance formation due to climate change. Satellite-based measurements provide a promising approach to estimate ground-level air pollution on large scale. However, most applications of satellite-based measurements have been used for fine particulate matter and nitrogen dioxide, while only a few have been used for O 3. In this study, we incorporated satellite-based measurements from the Ozone Monitoring Instrument (OMI) and MOderate-resolution Imaging Spectroradiometer (MODIS) with meteorological variables and land-use data to estimate daily maximum 8-h average O 3 at 1-km resolution in Taiwan during 2004–2020. The random forest model was used to impute the missing values of the satellite-based measurements. Additionally, the XGBoost model was leveraged to estimate daily O 3 concentrations. Model performance was evaluated by the ten-fold cross-validation (CV), temporal and spatial validation, and the results were reported as the coefficient of determination (R 2 ) and root mean square error (RMSE). Our results showed that the 10-fold CV, temporal validated, and spatial validated R 2 (RMSE) of the XGBoost model were 0.82 (7.71 ppb), 0.63 (11.09 ppb), and 0.68 (10.27 ppb), respectively. Our model performance was better in central and southern Taiwan. The top ten important predictors were date (relative importance = 12.15%), temperature (10.77%), meridional wind (10.71%), relative humidity (9.60%), zonal wind (8.14%), UV radiation (8.07%), total precipitation (6.35%), surface pressure (5.34%), surface O 3 volume mixing ratio (4.93%), and boundary layer height (4.69%). The spatial distribution of O 3 estimates showed that daily maximum 8-h average O 3 concentrations were higher in the suburban and mountainous areas near the central and southern Taiwan. This reveals that sensitive populations should still pay attention to the secondary pollutants even when outside the urban areas. The O 3 estimates can be further leveraged to evaluate the short-term and long-term effects of O 3 on human health. [Display omitted] • An estimation model for DM8O3 at 1-km resolution during 2004–2020 was developed. • OMI and MODIS products were integrated in the model. • A stronger correlation between DM8O3 and surface volume mixing ratio was observed. • Cross-validated R 2 and RMSE of the XGBoost were 0.82 and 7.71 ppb, respectively. • Estimated DM8O3 were higher in the suburban and mountainous areas in Taiwan. [ABSTRACT FROM AUTHOR]
Titel: |
A machine learning model for estimating daily maximum 8-hour average ozone concentrations using OMI and MODIS products.
|
---|---|
Autor/in / Beteiligte Person: | Jung, Chau-Ren ; Chen, Wei ; Chen, Wei-Ting ; Su, Shih-Hao ; Chen, Bo-Ting ; Chang, Ling ; Hwang, Bing-Fang |
Zeitschrift: | Atmospheric Environment, Jg. 331 (2024-08-15), S. N.PAG |
Veröffentlichung: | 2024 |
Medientyp: | academicJournal |
ISSN: | 1352-2310 (print) |
DOI: | 10.1016/j.atmosenv.2024.120587 |
Schlagwort: |
|
Sonstiges: |
|